Sign Language Recognition

Authors:

Zlata Stefanović, Mathematical Grammar School, Belgrade

Vladan Bašić, Gymnasium, Kraljevo

Mentor:

Aleksa Račić, School of Electrical Engineering, University of Belgrade

Abstract

This paper presents a system for recognizing American Sign Language (ASL) signs from hand images. Various classification approaches were implemented, including k-Nearest Neighbours (kNN), Keypoint Classification, Convolutional Neural Networks (CNN), and the advanced deep learning model EfficientNetB3. The data used came from a synthetically generated ASL sign dataset. Several preprocessing and augmentation techniques were applied to improve model robustness. The achieved accuracy ranged from 56% to 99%, with EfficientNetB3 yielding the best results.

Full Paper